What do you like best about Databricks?
1) In our implementation, Genie Space is actively used to enable NLQ-based access across multiple data products like Finance, HR, Marketing, Sales, and Supply Chain (inventory, demand planning, and replenishment), reducing dependency on data teams for ad-hoc queries.
2) We designed separate Genie Spaces for each BU/team/data product, ensuring domain-level isolation while still supporting cross-functional querying where required (e.g., Finance + Sales joins).
Each Genie Space is carefully configured with curated data tables, business-level instructions, and semantic context, which significantly improves the accuracy of SQL generation.
3) We provide few-shot examples, guided prompts, and sample business questions tailored to each domain, helping Genie understand real business intent instead of generic query patterns.
4) In Chat Mode, business users directly ask questions in natural language, and Genie translates them into SQL and returns results, which has improved self-service analytics adoption.
5) In Agent Mode, Genie goes beyond SQL generation by creating a logical execution plan, breaking down complex queries into multiple steps before querying the underlying data.
6) We built a dedicated Anomaly Detector Genie Space, where users ask questions about cluster cost, performance issues, and inefficient workloads.
This anomaly-focused Genie analyzes long-running jobs, inefficient queries, and cluster utilization patterns, using historical workload data to identify optimization opportunities.
7) A key implementation is notebook-level analysis, where Genie highlights code issues, shows before vs after optimization, categorizes problems (performance, cost, inefficiency), and explains improvements clearly.
8) Genie also provides quantified recommendations, including expected cost savings (e.g., idle cluster reduction, query tuning impact) and workload-based optimization strategies, making it highly actionable for engineering teams.
9) We extended Genie into Genie Code integrated with Databricks AI Assistant, enabling an agentic development experience directly within our data engineering workflows.
Our team defined custom skills in Markdown (MD files) such as Coder, Tester, Mapper, and Data Generator, which are attached to Genie Code to modularize capabilities.
These skills are used to support end-to-end SDLC activities, including code generation, transformation logic creation, test case design, and synthetic data generation.
10) Genie Code operates by first creating a structured execution plan, outlining all required steps before starting any development activity.
It then breaks the plan into a detailed to-do list, executing each step sequentially (e.g., create notebook → write transformation → validate logic → optimize code).
11) During execution, Genie Code follows a human-in-the-loop model, asking for approvals at every step with options like allow once, always allow, or read-only execution.
The behavior of Genie Code is controlled through project-specific guidelines and instructions, ensuring it aligns with our coding standards, architecture patterns, and governance rules.
12) It acts as a co-developer within the workspace, assisting engineers in writing optimized code, validating logic, and ensuring best practices are followed consistently.
We are leveraging it for proactive development workflows, where Genie not only executes tasks but also suggests improvements and optimization opportunities during development itself.
This approach has enabled a “vibe coding” style of development, where engineers focus on intent while Genie handles structured execution, resulting in faster delivery, reduced manual effort, and improved overall code quality. Review collected by and hosted on G2.com.
